Teardown No. 116 · Paid ads

10,000 Keywords, 3 Kept: How Andre Vilar handed his Amazon book ads to an AI agent and is losing money half as fast

Andre Vilar publishes books on Amazon on nights and weekends, alongside a corporate day job. His first four books found no readers, and running Amazon ads himself cost $56 to sell one $15 book. So he built an AI agent to run the ads, with a US$600 monthly budget and guardrails. It screened over 10,000 keywords for each book, and in one case kept 3. It scores every live search term, cuts the noise and only scales what converts. It also handles pricing and support escalations, and he approves its decisions. From Oct 1 to 24 his books earned $188 in royalties on $363 of ads, a loss of $174, against a loss of $346 over the same days in September.

Try this step

Stop learning ads and hand them to an agent. "So I stopped trying to become a media buyer and built an agent that does it for me." It has a US$600 monthly budget and "a series of guardrails (this time)."

The problem

“Put books on Amazon. nobody found them,” he writes. He published four books before book five brought the first sale, in month five. Amazon’s answer was to run ads, and he had never run one. His first try spent $56 to sell one $15 book. Then he changed a price and his ads froze for six weeks: $0 spent and 0 sold. In April he set up ads by hand with Claude’s help and left for a vacation. After 20 days he had sold over 15 books and spent twice the revenue. He could see sales starting, but not a way to scale them.

The blueprint (as he describes it)

  1. Stop learning ads and hand them to an agent. “So I stopped trying to become a media buyer and built an agent that does it for me.” It has a US$600 monthly budget and “a series of guardrails (this time).”
  2. Screen keywords at scale. The agent screened 10,000+ keywords for each book. In his Oct 22 post: “gave one 10,000 keywords. it kept 3.”
  3. Score every live search term and cut the noise. It scores every live search term, kills the noise and only scales what converts. On its last run it flagged $89 of $687 in spend.
  4. Test first, then keep, fix or kill. In September he gave the agent a budget to test which ads to keep, improve or kill. It killed one campaign running at 213% ACOS.
  5. Let it touch pricing too. The agent runs ads, pricing and support escalations. Repricing raised the royalty per book by 36%.
  6. A person approves. “I approve decisions and write the checks.”

Stated results

  • Sept 1-24: $402 on ads and $56 in royalties, a loss of $346. Oct 1-24: $363 on ads and $188 in royalties, a loss of $174. “Still losing money. Half as fast, spending less, earning 3.4x more.”
  • Month 2 of the agent: “My AI-run publishing house lost $146 this month.” The month before, it lost $424 (Oct 23 post).
  • What helped most, in his words, “wasn’t ads. It was repricing (+36% royalty per book) and killing one campaign bleeding at 213% ACOS.”

Caveats

  • It is still losing money. Two months of data, small numbers, and no audit beyond his own screenshots.
  • His two October loss figures, posted a day apart, differ ($146 on Oct 23 for “this month”, $174 for Oct 1-24 on Oct 24). The posts don’t explain the gap.
  • He doesn’t name the books, the model behind the agent or how it connects to Amazon Ads.
  • Repricing, not the ads themselves, is what he credits most.
  • The next bottleneck he names is reviews: 3 against competitors’ 6,000.
  • His stated phase 2 is turning the agent into “an automated CMO for indie authors,” so he may sell it later.
  • Handles: X @AndreDVilar (bio: “real numbers of my AI publishing business”, linking his Amazon book). The book page credits A.D Vilar. No Bluesky, Threads, Instagram, TikTok or LinkedIn profile verified.

FAQ

How does Andre Vilar use an AI agent for Amazon book ads?

The agent has a US$600 monthly budget and guardrails. It screens thousands of keywords per book, scores every live search term, cuts wasted spend and scales only what converts, and he approves its decisions.

Is it profitable?

No. From Oct 1 to 24 his books earned $188 in royalties on $363 of ads, a $174 loss, against a $346 loss over the same days in September.

What made the biggest difference?

In his words, repricing (+36% royalty per book) and killing one campaign running at 213% ACOS, not the ads themselves.

What is holding the books back now?

Reviews. He has 3, against competitors’ 6,000.

Sources


Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.